The affine invariant of generalized semitoric systems

THE AFFINE INVARIANT OF GENERALIZED SEMITORIC SYSTEMS

arXiv:1307.7516v1 [math.SG] 29 Jul 2013

´ ALVARO PELAYO

TUDOR S. RATIU

˜ NGO SAN VU .C

Abstract. A generalized semitoric system F := (J, H) : M → R2 on a symplectic 4-manifold is an integrable system whose essential properties are that F is a proper map, its set of regular values is connected, J generates an S 1 -action and is not necessarily proper. These systems can exhibit focusfocus singularities, which correspond to fibers of F which are topologically multi-pinched tori. The image F (M ) is a singular affine manifold which contains a distinguished set of isolated points in its interior: the focus-focus values {(xi , yi )} of F . By performing a vertical cutting procedure along the lines {x := xi }, we construct a homeomorphism f : F (M ) → f (F (M )), which restricts to an affine diffeomorphism away from these vertical lines, and generalizes a construction of V˜ u Ngo.c. The set ∆ := f (F (M )) ⊂ R2 is a symplectic invariant of (M, ω, F ), which encodes the affine structure of F . Moreover, ∆ may be described as a countable union of planar regions of four distinct types, where each type is defined as the region bounded between the graphs of two functions with various properties (piecewise linear, continuous, convex, etc). If F is a toric system, ∆ is a convex polygon (as proven by Atiyah and Guillemin-Sternberg) and f is the identity.

1. Introduction Let (M, ω) be a symplectic 2n-manifold, that is, M is a smooth 2n-dimensional manifold M and ω is a non-degenerate closed 2-form. Throughout this paper we assume that M is connected. However, we do not assume that M is compact. 1.1. Definitions. Motivated by [At82, GS82, PRV12, PV09, PV11, Vu07], we introduce in this paper a particular class of classical Liouville integrable systems of the so-called “generalized semitoric type”. Definition 1.1 An integrable system on (M, ω) is given by a map F : M → Rn whose components f1 , . . . , fn : M → R are Poisson commuting smooth functions which generate vector fields Xf1 , . . . , Xfn (via pairing with ω) that are linearly independent at almost every point. A singularity of F is a point in M where this linear independence fails to hold. A singular fiber of F is a level set of F which contains at least one singularity of F . In this article we assume that n = 2, and use the index free notation f1 = J and f2 = H. Definition 1.2 An S 1 -action on (M, ω) is Hamiltonian if there exists a smooth map J : M → R, the momentum map, such that ω(XM , ·) = −dJ, 1

´ ALVARO PELAYO

2

TUDOR S. RATIU

˜ NGO SAN VU .C

. Figure 1.1. The singular Lagrangian fibration F : M → R2 of a generalized semitoric system with three isolated singular values c1 , c2 , c3 . The generic fiber is a 2-dimensional torus, the singular fibers are lower dimensional tori, points, or multipinched tori. For each ~ ∈ {−1, 1}2 we construct, in Theorems B and C, a homeomorphism f~ : F (M ) → R2 such that (f~ ◦ F )(M ) is a “nice region” of R2 , which is a symplectic invariant. The notion of “nice region” is made precise in Definition 4.2. where XM is the infinitesimal generator of the action. In this article we construct a symplectic invariant when F is of generalized semitoric type. We refer to Section 8.2 for a quick review of the notions concerning singularities used in the following definition. Definition 1.3 An integrable system F := (J, H) : M → R2 on (M, ω) is generalized semitoric if: (H.i) (H.ii) (H.iii) (H.iv)

J is the momentum map of an effective Hamiltonian circle action. The singularities of F are non-degenerate with no hyperbolic blocks. F is a proper map (i.e., the preimages of compact sets are compact). J has connected fibers, and the bifurcation set of J is discrete (here discrete includes multiplicity: that is, for any critical value x of J, there exists a small neighborhood V 3 x such that the critical set of J in the preimage J −1 (V ) only contains a finite number of connected components.)

Remark 1.4 (H.iv) implies that the fibers of F are also connected by [PRV12]. (H.iii), (H.iv) are implied by (H.i),(H.ii) when J is proper. In some simple physical models like the spherical pendulum (Example 6.1), J is not proper but (H.iii), (H.iv) still hold. A typical generalized semitoric system is depicted in Figure 1.1. For background material on integrable systems and group actions, see [PV11a].


3

1.2. Singularities. The class of systems in Definition 1.3 may have the so called focus-focus singularities, and give rise to fibers of F which are multi-pinched tori. Focus-focus singularities appear in algebraic geometry [GS06] and symplectic topology, e.g., [LS10, Sy01, Vi2013] (in the context of Lefschetz fibrations they are sometimes called nodes), and include simple physical models from mechanics such as the spherical pendulum ([AM78]).

Figure 1.2. In general F (M ) ⊆ R2 is not convex. The interior of F (M ) contains two isolated singular values c1 = (x1 , y1 ) and c2 = (x2 , y2 ). By performing a vertical cutting procedure along the lines ì := {x := xi }, we construct a homeomorphism f : F (M ) → f (F (M )), which restricts to an affine diffeomorphism away from these vertical lines. The right hand side figure displays the associated polygon with the distinguished lines.

type I

type II

type IV

type III

type I

type II

Figure 1.3. A cartographic projection of F . It is a symplectic invariant of F , see Theorem C.

1.3. Toric systems: Atiyah and Guillemin–Sternberg Theory. If M is compact and F := (J, H) is the momentum map of an effective Hamiltonian 2torus action, all assumptions above hold by the Atiyah and Guillemin-Sternberg

4

´ ALVARO PELAYO

TUDOR S. RATIU

˜ NGO SAN VU .C

Theorem ([At82], [GS82]) and F does not possess any focus-focus singularity. In this case (M, ω, F ) is called a toric system or a symplectic toric manifold. Toric systems have been thoroughly studied in the past thirty years (in any dimension) and, at least from the point of view of symplectic geometry, a complete picture emerged in the compact case due to the aforementioned results of Atiyah [At82], Guillemin-Sternberg [GS82], and a classification result due to Delzant [De88]. The first two papers showed that the image µ(M ) of the momentum map µ : M → Rk of a Hamiltonian Tk -action on a 2n-dimensional compact connected symplectic manifold (M, ω) is a convex polytope in Rk , which is a symplectic invariant. 1.4. Goal of this article. In the present article we will extend to generalized semitoric systems the results in Section 1.3, inspired by an extension of the Atiyah-Guillemin-Sternberg result to (non-generalized) semitoric systems recently achieved by V˜ u Ngo.c [Vu07]. Using Morse theory and the Duistermaat-Heckman Theorem for proper momentum maps, he dealt with integrable systems F : M → R2 of semitoric type for which, in addition to assumptions (H.i)-(H.iv), J : M → R is proper. Then he performed a cutting procedure along the vertical lines going through the isolated singularities of the image F (M ) of the system and constructed a convex polygon from it, which is an invariant of F ; see Figure 1.2. The difficulty of the generalized situation considered in this article is due to the fact that the Duistermaat-Heckman theorem does not hold for nonproper J (Remark 4.4), and neither does standard Morse theory. This has striking consequences for the statement of our extension: while the invariant in [Vu07] is a class of convex polygons as in Figure 1.2, ours is a union of planar regions of various types (to be precisely defined later), which looks, in general, like Figure 1.3. This invariant encodes the singular affine structure induced by the (singular) Lagrangian fibration F : M → R2 on the base F (M ). Its construction and properties appear in Theorems B, C, D. This affine structure also plays a role in parts of symplectic topology, mirror symmetry, and algebraic geometry, see for instance Auroux [Au09], Borman-Li-Wu [BLW13], Kontsevich-Soibelman [KS06]. Integrable systems exhibiting semitoric features appear in the theory of symplectic quasi-states, see Eliashberg-Polterovich [EP10]. Theorem E shows that there are many simple examples in which the invariant, which is the most natural planar representation of the singular affine structure of the system, has a non-polygonal, non-convex, form. 2. Toric and semitoric systems Although this section is not original, we put previous results in a general framework which is better suited for expressing our new results in the following section. 2.1. The set of semitoric images. Let P(R2 ) be the set of subsets of R2 and n o F := Z, Z+ , Z− , {1, . . . , N }N >0 , ∅ ,


5

where Z+ = {n ∈ Z | n > 0} and Z− = {n ∈ Z | n 6 0}. Let 1 0 T := , 1 1

and consider the group T whose elements are the matrices T k , k ∈ Z, composed with vertical translations. This gives rise to the quotient space P T (R2 ) := P(R2 )/T . 2.2. Action on P T (R2 ) × RZ × NZ . A vertical line L ⊂ R2 splits R2 into two half-spaces. Let u ∈ Z. We define a map tuL acting on R2 as follows. On the left half space defined by L, we let the map tuL act as the identity. On the right half space, with an origin placed arbitrarily on L, tuL acts as the matrix T u . Definition 2.1 Let Z ∈ F and let ~x ∈ RZ . Let n ∈ Z and denote by L~nx the vertical line through (~x(i), 0). We define the action of ~u ∈ ZZ on P T (R2 ) × RZ by    Y ~ u(i) ~u · (X, ~x) =  tL~x  (X), ~x . n

~ u(i)6=0, n∈Z

Let ~k ∈ NZ . We finally define the action of ~ ∈ {−1, 1}Z on P T (R2 ) × RZ × NZ by the formula ~ ~ ~ ~ · (X, ~x, k) = (~ · k) · (X, ~x), k ,

Z where ~·~k := i 7→ 1−(i) 2 k(i). We denote the {−1, 1} -orbit space by BGST (Z) := P T (R2 ) × RZ × NZ /{−1, 1}Z .

2.3. Affine invariant for semitoric systems. Let F = (J, H) : M → R2 be a semitoric system, i.e., in addition to assumptions (H.i)-(H.iv), the map J : M → R is proper. There exists a unique Z ∈ F such that ~x ∈ RZ is the tuple of images by J of focus-focus values ci = (xi , yi ) of F ordered by non-decreasing values, and ~k ∈ NZ such that ~k(i) is the number of focus-focus critical points in the fiber F −1 (ci ). Let L~x := (Lxi )i∈Z where Lxi is the unique vertical line in R2 through (xi , 0). For each fixed ~ ∈ {−1, 1}Z , V˜ u Ngo.c constructed [Vu07, Theorem 3.8 and Proposition 4.1] an equivalence class of convex polygons in R2 (2.1)

(∆~ mod T ) ∈ P T (R2 ).

by performing a cutting procedure along the vertical lines Lxi . The “choice” of cuts is given by ~, where a positive sign corresponds to an upward cut, and a negative sign corresponds to a downward cut. Definition 2.2 Let (M, ω, F ) be a semitoric system. Define: (2.2) ∆(M, ω, F ) := (∆~ mod T , ~x, ~k) mod {−1, 1}Z ∈ BGST ,

where ~(i) = 1 for all i ∈ Z and the action of {−1, 1}Z is defined above. Definition 2.3 Let MT be the set of toric systems. Let Z ∈ F and let MST (Z) and MGST (Z) be the sets of semitoric and generalized semitoric systems F =

´ ALVARO PELAYO

6

TUDOR S. RATIU

˜ NGO SAN VU .C

(J, H), with the images of the focus-focus values of F by J indexed by the set Z. Let G G G MST := MST (Z); MGST := MGST (Z); BGST := BGST (Z). Z∈F

Z∈F

Z∈F

We now recall the notion of isomorphism for generalized semitoric systems, which coincide with the notion introduced in [Vu07] for proper semitoric systems. Definition 2.4 The generalized semitoric systems (M1 , ω1 , F1 := (J1 , H1 )) and (M2 , ω2 , F2 := (J2 , H2 )) are isomorphic if there exists a symplectomorphism ϕ : M1 → M2 such that ϕ∗ (J2 , H2 ) = (J1 , h(J1 , H1 )) for a smooth h ∂h > 0. such that ∂H 1 Notice that the set MT is not invariant under these isomorphisms. Hence we introduce the following definition. Definition 2.5 A generalized semitoric system is said to be of toric type if it is isomorphic to a toric system. We denote by MTT the set of semitoric systems of toric type. Remark 2.6 Clearly MT ( MST ( MGST .

Theorem 2.7 ([Vu07]). The class of convex polygons (2.2) is an invariant of the isomorphism type of F .1 For a system satisfying properties (Hi)–(Hiv), in this article we will construct a more general symplectic invariant by unwinding the (singular) affine structure induced by F on F (M ), which extends (2.1). The fact that J may not be proper complicates the situation a lot because the Duistermaat-Heckman theorem does not hold for nonproper momentum maps (Remark 4.4), and standard Morse theory essentially breaks down for nonproper maps. This is why systems with non-proper J were excluded in [PV09, PV11]. 3. Summary result: Theorem A As a consequence of Theorems B, C, and D (stated and proved in the next sections), we obtain the following statement, which is less explicit (less useful for computations) but provides a summary of the paper. Definition 3.1 Recall that if (M, ω, F ) ∈ MT then F (M ) does not contain focus-focus singular values, and F (M ) is a convex polygon. If (M, ω, F ) ∈ MST , let ∆(M, ω, F ) be as in (2.2). Consider the maps (3.1)

CST : MST 3 (M, ω, F ) 7−→ ∆(M, ω, F ) ∈ BGST

(3.2) CT : MT 3 (M, ω, F ) 7−→ (F (M ) mod T , ∅, ∅) 1while F (M ) is neither generally convex, nor an invariant.

mod {−1, 1}Z ∈ BGST


and (3.3) CTT : MTT 3 (M, ω, F ) 7−→ (F 0 (M ) mod T , ∅, ∅)

7

mod {−1, 1}Z ∈ BGST ,

where F 0 is any toric momentum map isomorphic to F as a semitoric system. Definition 3.2 If F is a family of integrable systems containing MT , a cartographic invariant is any map C : F → BGST extending CT in (3.3) and invariant under isomorphism. It follows from the Atiyah-Guillemin-Sternberg theory and [Vu07] that the maps CT , CTT and CST are cartographic invariants. Notice that it is straightforward to check from Definition 2.4 that CTT is indeed well defined. Theorem A. Let CST , CTT and CT be the cartographic invariants defined in (3.1) and (3.3). Then there exists a cartographic invariant CGST : MGST → BGST such that the diagram (3.4)

MT

/ MTT

/ MST CTT

/ MGST

CST

+) $

CT

CGST

BGST

is commutative. We will prove several theorems which together imply Theorem A and which are more informative because the cartographic invariant is explicitly constructed. It would be interesting to prove Theorem A (in particular, defining the maps involved) for integrable systems on origami manifolds (see [DGP11]) and on orbifolds (see [LT97]), where, as far as we know, integrable systems have not been studied. 4. Main results: Theorems B, C, D, E For simplicity, from now on, we use the term “semitoric” to refer to integrable systems satisfying (Hi)–(Hiv), that is, we drop the word “generalized”. Let (M, ω) be a connected symplectic 4-manifold and F := (J, H) : M → R2 a semitoric system. Next we prepare the grounds for the main theorems of the paper. Let Br ⊂ B is the set of regular values of F . Since F is proper we know that the set of focus-focus critical values of F is discrete. We denote by ci := (xi , yi ), i ∈ Z, the focus-focus critical values of F , ordered so that xi 6 xi+1 , and ki is the number of critical points in F −1 (ci ). Given ~ = (i )i∈Z ∈ {−1, +1}Z , we define the vertical closed half line originating at ci = (xi , yi ) by Li i := {(xi , y) ∈ R2 | i y > i yi }

for each i ∈ Z, which is pointing up from ci if i = 1 and down if i = −1. Define ì i := B ∩ Li i ⊂ R2 . For any c ∈ B, define Ic := {i ∈ Z | c ∈ ì i } and the map k : R2 → Z by X (4.1) k(c) := i ki , i∈Ic

´ ALVARO PELAYO

8

TUDOR S. RATIU

˜ NGO SAN VU .C

with the convention that if Ic = ∅ then k(c) = 0. The sum is finite thanks to (H.iv). Let `~ := k −1 (Z \ {0}). For the necessary background on affine manifolds in the discussion which follows, readers may consult the appendix (Section 8.3). We write A2Z for R2 equipped with its standard integral affine structure with automorphism group Aff(2, Z) := GL(2, Z) n R2 . The integral affine structure on Br , which in general is not the affine structure induced by A2Z , is defined for instance in [Vu07, Section 3] or [HZ1994, Appendix A2]; see also Section 8.3: affine charts near regular values are given by action variables f : U ⊂ F (M ) → R2 on open subsets U of F (M ) with the induced subspace topology and any two such charts differ by the action of Aff(2, Z). Let X and Y be smooth manifolds and A ⊂ X. A map f : A → Y is said to be smooth if every point in A admits an open neighborhood in X on which f can be smoothly extended. The map f is called a diffeomorphism onto its image if f is injective, smooth, and its inverse f −1 : f (A) → A is a smooth map, in the sense above. The following theorem is a generalization of [Vu07, Theorem 3.8]. Theorem B. Let F : M → R2 be a semitoric system in MGST (Z), for some Z ∈ F. For every ~ ∈ {−1, +1}Z there exists a homeomorphism f~ : B → f~(B) ⊆ R2

(2)

of the form f~(x, y) = (x, f~ (x, y)) such that: (P.i) the restriction f~|(B\`~) is a diffeomorphism onto its image, with positive Jacobian determinant; (P.ii) the restriction f~|(Br \`~) sends the integral affine structure of Br to the

standard integral affine structure of A2Z ; (P.iii) the restriction f~|(Br \`~) extends to a smooth multi-valued map Br → R2 and for any i ∈ Z and c ∈ ì i \ {ci }, we have

(4.2)

lim df~(x, y) = T k(c) lim df~(x, y),

(x,y)→c xxi

where k(c) is defined in (4.1). Such an f~ is unique modulo a left composition by a transformation in T . In toric case, f~(x, y) = (x, y), as was mentioned in Section 3. Definition 4.1 The map f~ in Theorem B is a cartographic map2 for F and its image f~(B) is a cartographic projection of F . Definition 4.2 Let R be a subset of R2 . We say that R has type I if there is a convex polygon ∆ ⊂ R2 and an interval I ⊆ R such that n o R = ∆ ∩ (x, y) ∈ R2 | x ∈ I . 2since they lay out the affine structure of F on two dimensions.


9

We say that R has type II if there is an interval I ⊆ R and f : I → R, g : I → R such that f is piecewise linear, continuous, and convex, g is lower semicontinuous, and n o R = (x, y) ∈ R2 | x ∈ I and f (x) 6 y < g(x) .

We say that R has type III if there is an interval I ⊆ R and f : I → R, g : I → R such that f is upper semicontinuous, g is piecewise linear continuous and concave, and n o R = (x, y) ∈ R2 | x ∈ I and f (x) < y 6 g(x) . We say that R has type IV if there is an interval I ⊆ R and f, g : I → R such that f is upper semicontinuous, g is lower semicontinuous, and n o 2 R = (x, y) ∈ R | x ∈ I and f (x) < y < g(x) .

In the following statement we call a discrete sequence a sequence such that for every value c, there is a neighborhood of c which contains only the image of a finite number of indices. Theorem C. Let F = (J, H) : M → R2 be a semitoric system and let f~ be a cartographic map for F . Let n o K + := x ∈ J(M ) | J −1 (x) ∩ H −1 ([0, +∞)) is compact . and

n o K − := x ∈ J(M ) | J −1 (x) ∩ H −1 ((−∞, 0]) is compact .

Suppose that the topological boundaries ∂K + and ∂K − in J(M ) are discrete. Then there exists an increasing sequence (xj )j∈Z in R, and sets Cj~ ⊂ R2 , j ∈ Z, such that: (P.1) for each j ∈ Z, the set Cj~ has type I, II, III, or IV associated to (xj , xj+1 ); S (P.2) f~(B) = j∈Z Cj~; (P.3) for every j ∈ Z, and every regular value x of J, the volume V (x) 6 +∞ of J −1 (x) is equal to the Euclidean length of the vertical line segment ({x} × R) ∩ Cj~. In some cases (for instance if M is compact), only a finite number of the xj ’s are relevant. Suppose that F : M → R2 is the momentum map of a Hamiltonian T2 action on a compact connected symplectic 4-manifold. Then the cartographic projection of F is a compact convex polygon in R2 ; see [At82] and [GS82]. If F : M → R2 is a semitoric system for which J is proper, then any cartographic projection of F is a convex polygon in R2 , which may be bounded or unbounded, and which is always a closed subset of R2 ; see [Vu07, Theorem 3.8]. Example 4.3 Figure 1.1 shows the regularSand singular focus-focus fibers of singular Lagrangian fibration f ◦ F : M → j∈Z Cj~ in Theorem C. There are two focus-focus singular fibers, F −1 (ci ), i = 1, 2. The value c1 has multiplicity k1 = 2 and c2 has multiplicity k2 = 3.

´ ALVARO PELAYO

10

TUDOR S. RATIU

˜ NGO SAN VU .C

Remark 4.4 Concerning Theorem C(P.3), note that the Duistermaat-Heckman theorem does not hold for nonproper momentum maps. Indeed, let M = S 2 ×S 2 with F = (z1 , z2 ) (toric momentum map). Let f : [−1, 1] → (−1, 1] be continuous. Let M 0 = F −1 ({(x, y) | x ∈ [−1, 1], y < f (x)}. The set M 0 is an open subset of M , and µ = z1 is a momentum map for a Hamiltonian S 1 -action on M 0 . Furthermore, µ is not proper because µ−1 (x) = F −1 ({(x, y) | y < f (x)}) is not closed. Now let V (x) be the symplectic volume of Mx0 where Mx0 = M 0 ∩ µ−1 (x)/S 1 = S 2 ∩ {z2 < f (x)}. (See Figure 4.1.) Then V (x) = vol(S 2 ) [(1 + f (x))/2] = 2π(1 + f (x)). So V (x) is not piecewise linear in general, in contrast with Duistermaat-Heckman [DH82]. This shows that the Duistermaat-Heckman theorem may not hold when the S 1 -momentum map is not proper. Notice that the full map FM is not proper, but we can easily modify it as follows. Let g(x, y) = (x, 1/(f (x) − y)). Then F 0 = g ◦ F |M 0 is proper, and the S 1 -momentum map is not modified. Thus F 0 is a generalized semitoric system. z2

1111111111111 0000000000000 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111

000000000000000000 111111111111111111 111111111111111111 000000000000000000 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 x2 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111

f (x)

y2

Figure 4.1. The reduced manifold Mx0 . In the following definition, we use the terminology of sections 2 and 3. Definition 4.5 Let CGST (F ) be defined as follows. Let ~ = (i )i∈Z with i = 1 for all i ∈ Z. Then

(4.3)

CGST (F ) := (f~(B) mod T )

mod {−1, 1}Z ∈ BGST .

Theorem D. The map CGST : MGST → BGST is a cartographic invariant. Proof. Let F1 : M1 → R2 and F2 : M2 → R2 be semitoric systems, and let f~,1 , f~,2 be the corresponding cartographic maps defined by Theorem B. If F1 and F2 are isomorphic, they have the same leaf space, with identical induced integral affine structures. Thus, from Theorem B, (P.ii), there must be a transformation t ∈ T such that f~,1 = t · f~,2 . Then the result follows from (4.3). We conclude with a result which shows that there are semitoric systems with a cartographic projection which may not occur as the cartographic projection of a toric or semitoric system (J, H) : M → R2 with proper J.


11

Theorem E. There exists an uncountable family of semitoric integrable systems {Fλ : M → R2 }λ , with cartographic map fλ,~ , such that the following properties hold: (E.1) (E.2) (E.3) (E.4) (E.5) (E.6) (E.7)

Fλ : M → R2 is proper; Bλ := Fλ (M ) is unbounded in R2 ; fλ,~(Bλ ) is a bounded in R2 ; Bλ is not a convex region; Bλ is not open and is not closed in R2 ; Fλ is isomorphic to Fλ0 if and only of λ = λ0 ; for every i ∈ {I, II, III, IV } there exists λ such that fλ,~(B) as in Theorem C(P.ii), is a union of regions in R2 of types I, II, III, and IV, in which at least one of them has type i.

Motivated by [PV11], the following inverse type question is natural. Let C := ∪j∈N Cj be a connected set, where Cj ⊂ R2 is a region of type I, II, III, or IV. Does there exist a semitoric system F : M → R2 with B := F (M ) such that f~(B) = C, where f~ is a cartographic map for F ? The classifications of Delzant [De88] and [PV11] give partial answers to this question. Note that here we are not claiming uniqueness; in fact, it follows from [PV11] that there are many semitoric systems which realize the same C. 5. Proof of Theorem B The proof is close to [Vu07], but our construction is more transparent thanks to the use of a recent result in [PRV12]. Let ΣJ be the bifurcation set of J. We fix a point q0 = (x0 , y0 ) ∈ Br , such that x0 ∈ / ΣJ . Since the fibers of J are connected by (H.iv), we know from [PRV12, Theorem 4.7] that the fibers of F are also connected. By the Liouville-Mineur-Arnold Theorem (see, [HZ1994, Appendix A2]), there exists a diffeomorphism g : U ⊂ Br → g(U ) ⊂ R2 , with positive Jacobian determinant, defined on an open neighborhood U of q0 which, without loss of generality, we may assume to be simply connected, such that A = (A1 , A2 ) = g ◦ F are local action variables. Since J is the momentum map of an effective Hamiltonian S 1 -action, it has to be free on the regular fibers (see for instance [DK00, Theorem 2.8.5]). Hence, we may assume A1 = J (see [Vu07, point 2 of the proof of Theorem 3.8]). Repeating this argument with an open cover of Br , we may fix an affine atlas of Br such that all transition functions belong to the group T (see Section 2.1). We divide the proof into four steps: the first four treat the generic case in which the lines in `~ are pairwise distinct, whereas the last step deals with the non-generic case. We warn the reader that statements (P.i)–(P.iii) are proven in the first three steps, but the claim that f~ is a homeomorphism onto its open image is proven in Step 4.

The homeomorphism f~ with the required properties is constructed from the er → Br , chosen with q0 as base developing map of the universal cover pr : B 2 e : B er → R be the unique developing map such point (see Section 8.3). Let G

12

´ ALVARO PELAYO

TUDOR S. RATIU

˜ NGO SAN VU .C

e ([γ]) = g(γ(1)) for paths γ contained in U , such that γ(0) = q0 . The that G e in order to extend g to the whole image B = F (M ). goal is to use G

We begin the proof by assuming that the half lines in `~ do not overlap. Step 1. (Br \ `~ is simply connected ). By [PRV12, Theorem 4.7], B is a region of R2 which is between the graphs of two continuous functions defined on the same interval. These graphs cannot intersect above an interior point of the interval, because this would imply that the interior of Br is not connected, which is known to be false because F is proper (see [PRV12, Theorem 3.6]). This proves that Br \ `~ is simply connected.

Step 2. (Proof of (P.i) and (P.ii) on Br \ `~). Hence, the developing map e~ : B er → R2 induces a unique affine map G~ : Br \ `~ → R2 by the relation G e~, G~ ◦ pr := G

i.e., if c ∈ Br \ `~ and γ is a smooth path in Br \ `~ connecting q0 to c, then e~([γ]). Note that G~|U = g. G~(c) := G The definition implies that G~ is a local diffeomorphism. We show now that G~ is injective. Since A1 = J, G~|U is of the form G~(x, y) = (x, h~U (x, y)) for some smooth function h~U : U → R. Because we have an affine atlas of Br with transition functions in T , the affine map G~ must preserve the first component x, i.e. there exists a smooth function h~ : Br \ `~ → R, extending h~U such that G~(x, y) = (x, h~(x, y)) ~ for all (x, y) ∈ Br \ `~. Since G~ is a local diffeomorphism, ∂h ∂y never vanishes, which implies that for each fixed x, all the maps y 7→ h~(x, y) are injective. Hence G~ is injective and thus a global diffeomorphism Br \ `~ → G~ Br \ `~ ⊂ R2 . This proves (P.i) on Br \ `~ by choosing f~ := G~ and (P.ii) because G~ is an affine map. Step 3. (Extension of the developing map to B \ `~ and proof of (P.i) and (P.iii)). By the description of the image of F in [PRV12, Theorem 5], we simply need to extend G~ at elliptic critical values. But the behavior of the affine structure at an elliptic critical value c is well known (see [MZ04]): there exist a smooth map a : V → R2 , where V is an open neighborhood of c ∈ R2 , and a symplectomorphism ϕ : F −1 (V ) → MQ onto its image such that

(5.1)

a ◦ F |F −1 (V ) = Q ◦ ϕ : F −1 (V ) → R2 ,

where Q is the “normal form” of the same singularity type as F , given by Q = (x21 + ξ12 , ξ2 ) (rank 1 case) or Q = (x21 + ξ22 , x22 + ξ22 ) (rank 0 case). Here MQ = R2 × T∗ T1 = R2 × T1 × R (rank 1) or MQ = R4 (rank 0). It follows from the formula for Q that Q is generated by a Hamiltonian T2 -action, and therefore a is an affine map. On the other hand, since F and Q have the same singularity type, the ranks of dF and dQ must be equal, and the dimensions of the spaces spanned by the Hessians must be the same as well. Computing the Taylor expansion of (5.1) shows that da(c) has to be invertible. Thus, a


13

is a diffeomorphism onto its image. Therefore a|Br ∩V is a chart for the affine structure of Br . Thus there exists a unique affine map A ∈ Aff(2, Z) such that (G~)|Br∩V = A ◦ a|Br∩V

and we may simply extend G~ to Br ∪ V by letting (G~)|V = A ◦ a.

Because a is a diffeomorphism into its image, we see that G~ remains a local diffeomorphism. This proves (P.i) with f~|B\`~ := G~.

The fact that G~ extends to a smooth multi-valued map Br → R2 follows from the smoothness of the universal cover as in [Vu07, Section 3]. Formula (4.2) follows from the calculation of the monodromy around focus-focus singularities, which is carried out exactly as in [Vu07, pages 921-922] since it relies only on the properness of F (and not on the properness of J). This proves (P.iii). Step 4. (Extension to a homeomorphism B → R2 ). Finally we show that G~ may be extended to a homeomorphism f~ : B → f~(B) ⊂ R2 , which will prove the theorem if no half lines in `~ overlap. Because of (P.iii), if c0 ∈ `~, but c0 is not a focus-focus value, it follows that G~ has a unique continuation to c0 , from the left, and a unique continuation from the right. As in [Vu07, Proof of Theorem 3.8], the fact that these continuations coincide follows from the fact that the affine monodromy around a focus-focus singularity leaves the vertical line through c0 pointwise invariant. That G~(c) has a limit as c approaches the focus-focus value follows from the z logz behavior of G~, see [Vu03, Section 3]. Let f~ : B \ {ci | i ∈ Z} → R2 be this continuous extension of G~. Because of (P.iii), the extensions of the vertical derivative ∂y f~ from the left or from the right coincide on `~. Since any extension of G~(x, y) = (x, h~(x, y)) is a local diffeomorphism, ∂y h~ cannot vanish on `~. Thus, f~|`~ is injective. This implies that f~ is injective on B \ {ci | i ∈ Z}. Extend by continuity the map f~ to {ci | i ∈ Z}. So far, we have shown that f~ : B → R2 is a continuous injective map which is an affine diffeomorphism off `~. It remains to be shown that (f~)−1 is continuous on f~(B). Since f~ is a diffeomorphism off `~, we only have to show that (f~)−1 is continuous at points of f~(`~). b~ : U → G b~(U ) be an affine chart which coincides Let c0 = (x0 , y0 ) ∈ ˚ `~ and G with f~ on the left hand-side of c0 in U , that is, on n o Uleft := (x, y) ∈ U | x 6 x0 . Then,

b −1 |f (U ) (f~)−1 |f~(Uleft ) = G left ~ ~

and hence it is continuous on f~(Uleft ). Similarly, it is proved that (f~)−1 |f~(Uright ) is continuous on Uright , which shows that (f~)−1 is continuous at f~(c0 ) for any c0 ∈ ˚ `~.

14

´ ALVARO PELAYO

TUDOR S. RATIU

˜ NGO SAN VU .C

Finally, we need to prove the continuity of (f~)−1 at all points f~(ci ), where ci = (xi , yi ), i ∈ Z, are the focus-focus values in B. Let ì be the vertical line containing ci . Let us use the following local description of the behavior of f~ at ci , [Vu03], [Vu07, Proof of Theorem 3.8]: for all (x, y) ∈ U \ ì , f~(x, y) = (x, Re(z logz) + g(x, y)),

where z = yˆ(x, y) + ix ∈ C, g and yˆ are smooth functions and yˆ(0, 0) = y0 . It ~ follows that ∂f ∂y is continuous near ci (which is in agreement with (4.2)) and is equivalent, as z → 0, to K ln(x2 + y 2 ) for some constant K > 0. Hence we get the lower bound ∂f~ ∂y > C > 0 for some constant C, if (x, y) is in a small neighborhood V = [xi − η, xi + η] × [yi − η, yi + η] of ci , for some η > 0. For simplicity of notation, let us assume for instance that i = 1; the case i = −1 is treated similarly. Hence, for any fixed x ∈ [xi − η, xi + η], the function y 7→ f~(x, y) is invertible on (yi , yi + η] and has bounded derivative, uniformly for x ∈ [xi − η, xi + η]. Hence, the inverse (f~)−1 extends by continuity at f (ci ) = f (xi , yi ). The limit of the inverse at this point must equal yi since f~ is injective. This shows that (f~)−1 is continuous at the point f~(ci ). This concludes the proof of Theorem B in case there is no overlap of vertical lines in `~. Step 5. (Proof in the case of overlapping lines in `~). If on the other hand there are overlaps of vertical lines in `~, then Br \ `~ may not be simply connected. In this case, for each c ∈ Br \ `~, we need to choose a path γc joining q0 to c inside Br \ {ci | i ∈ Z}, which we do as follows. We replace the focus-focus critical values ci which lie in the same vertical line by nearby points e ci , in such a way that their x-coordinates are all pairwise distinct. This turns the corresponding set Br \ `˜~ into a simply connected set; thus, up to homotopy, there is a unique path γc joining q0 to c inside Br \ `˜~, and we can always assume that this path avoids the true focus-focus values ci . The homotopy class of γc depends on the choice of ordering of the x-coordinates of the points e ci . However, we claim that the value e~([γc ]) G~(c) := G is well defined. Indeed, decomposing a permutation as a product of transpositions of the form (i, i + 1) or (i + 1, i), it suffices to consider only the case where we permute two points, e ci and e ci+1 , which lie in adjacent vertical lines. In this case, one can check that the homotopy class [γc ] is modified by a commutator −1 gi gi+1 gi−1 gi+1 , where gi , i ∈ Z, is a set of generators of the fundamental group of Br \ {ci | i ∈ Z}. But the monodromy representation is Abelian, due to the e~([γc ]) global S 1 action (see [CVN02]). It follows that, as required, the value G is invariant under this transposition. Now that G~ is defined, the previous proof for (P.i) and (P.ii) remains valid. The formula in (P.iii) follows from the fact that the monodromy representation is Abelian.


15

6. Proof of Theorem C and the spherical pendulum example Proof of Theorem C. The proof is divided into three steps. Step 1. Let f~ : B → f~(B) ⊂ R2 be the homeomorphism in Theorem B. Let H + , H − : J(M ) → R be the functions defined by H + (x) := supJ −1 (x) H and H − (x) := inf J −1 (x) H. Since J is Morse-Bott with connected fibers (see, e.g., [PR11, Theorem 3]) we may apply [PRV12, Theorem 5.2] which states that H + , H − are continuous and F (M ) = (hypograph of H + ) ∩ (epigraph of H − ) . Since H + , H − are continuous and F is proper, one can check that the sets K + , K − defined in the theorem are open in J(M ). Hence we have the following equality of sets, where the four sets on the right hand side are open and disjoint: J(M ) = (K + ∩ K − ) ∪ (K + \ K − ) ∪ (K − \ K + ) ∪ (J(M ) \ (K + ∪ K − )). By assumption, ∂K + and ∂K − are discrete, and therefore there exists a countable collection of intervals {Ij }j∈Z , whose interiors are pairwise disjoint, such that − + − each Ij is contained in one of the above four sets (K + ∩ K S ), (K \ K ), − + + − (K \ K ) or (J(M ) \ (K ∪ K )), and such that J(M ) = j∈Z Ij . By letting for every j ∈ Z, Cj~ := f~((Ij × R) ∩ F (M )) ⊂ Ij × R, we obtain S f~(F (M )) = j∈Z Cj~,.

Step 2. (Proof of (P.1) and (P.2)). We consider the four cases.

(1) If Ij ⊂ (K + ∩ K − ), then the fibers of J are compact, and hence the analysis carried out in [Vu07, Theorem 3.8, (v)] applies. This implies that Cj~ is of type I. (2) Consider now Ij ⊂ (K − \K + ). Let x ∈ Ij . Since J −1 (x)∩H −1 ((−∞, 0]) is compact, H− (x) is finite. On the other hand, H+ (x) must be +∞; otherwise, F −1 ({x} × [0, H+ (x)]) would be compact, by the properness of F . This would imply that J −1 (x) is compact, a contradiction. (2) Let y ∈ H(J −1 (x)). Recall that f~(x, y) = (x, f~ (x, y)) and that (2)

∂f

∂f~ ∂y

(2)

~ is continuous on F (M ) (see (4.2)). Since ∂y f~((Ij × R) ∩ F (M )) = Cj~ has the form n o (x, z) | x ∈ Ij , h~− (x) 6 z < h~+ (x) ,

> 0, the image

where

h~− (x) := h~+ (x) :=

(2)

(2)

min

f~ (x, y) = f~ (x, H− (x)) ∈ R

sup

f~ (x, y) = lim f~ (x, y) ∈ R.

y∈J −1 (x)

y∈J −1 (x)

(2)

(2)

y→+∞

We have used the fact that f~ is a homeomorphism, so that the point (x, h~+ (x)) cannot belong to Cj~. The function h~+ is a pointwise limit of continuous functions, so it is continuous on a dense set. However, we need to show that it is lower semicontinuous. The new map (2) J, f~ (J, H) = f~ ◦ F

16

´ ALVARO PELAYO

TUDOR S. RATIU

˜ NGO SAN VU .C

satisfies the hypothesis of the following slight variation of [PRV12, Theorem 5.2] for continuous maps (the proof of which is identical line by line): c be a connected smooth four-manifold. Let Fb = (J, b H) b :M c → R2 Let M be a continuous map. Suppose that the component Jb is a smooth nonb+ b− constant Morse-Bott function with connected fibers. Let H , H : b + (x) := sup b−1 H c → R be defined by H b and H b − (x) := inf b−1 H. b Jb M J (x) J (x) b + and −H b − are lower semicontinuous. Then the functions H This statement gives the required semicontinuity in the statement of Theorem C. The analysis of the graph of h~− , which corresponds to the elliptic critical values and possible cuts due to focus-focus singularities, was carried out in [Vu07, Theorem 3.8]: it is continuous, piecewise linear, and convex. Thus, Cj~ is of type II. (3) The fact that Ij ⊂ (K + \ K − ) implies that Cj~ is of type III can be proved in a similar way to (2). (4) Finally, let Ij ⊂ J(M ) \ (K + ∪ K − ). In this case, we must have, for any x ∈ Ij , H+ (x) = +∞ and H− (x) = −∞. Therefore, f~((Ij × R) ∩ F (M )) = Cj~ has the form n o (2) (2) (x, z) | x ∈ Ij , lim f~ (x, y) < z < lim f~ (x, y) , y→−∞

y→+∞

where the limits are understood in R. Thus, Cj~ is of type IV.

This proves (P.1). Step 3. (Proof of (P.3)). By the action-angle theorem, (A1 , A2 ) := f~ ◦ F is a set of action variables near F −1 (x, y) with A1 = J,

A2 = A2 (J, H).

We have a symplectomorphism U → T2θ × R2A , where U is a saturated neighborhood of the fiber F −1 (x, y), and the symplectic form on T2θ × R2A is given by dA1 ∧ dθ1 + dA2 ∧ dθ2 . We have n o −1 2 −1 U ∩ J (x) = A1 (x) = (θ, A) | θ ∈ T , A1 = x . Since the normalized Liouville volume form is (2π)−2 dA1 ∧ dA2 ∧ dθ1 ∧ dθ2 , the induced volume form on U ∩J −1 (x) is (2π)−2 dA2 ∧dθ1 ∧dθ2 . In other words, the push-forward by A2 of the Liouville measure on J −1 (x) has a constant density 1 against the Lebesgue measure dA2 . This gives the result because the set of critical points of H in J −1 (x) has zero-measure in J −1 (x). This concludes the proof of Theorem C.

Example 6.1 (Spherical Pendulum) Semitoric systems with proper F = (J, H) but non-proper J include many simple integrable systems from classical mechanics, such as the spherical pendulum, which we now recall. The phase space of the spherical pendulum is M = T∗ S 2 with its natural exact symplectic form. Let the circle S 1 act on the sphere S 2 ⊂ R3 by rotations about the vertical axis. Identify T∗ S 2 with TS 2 , using the standard Riemannian metric on


17

S 2 , and denote its points by (q, p) = (q 1 , q 2 , q 3 , p1 , p2 , p3 ) ∈ T∗ S 2 = TS 2 , kqk2 = 1, q · p = 0. Working in units in which the mass of the pendulum and the gravitational acceleration are equal to one, the integrable system F := (J, H) : TS 2 → R2 is given by the momentum map of the (co)tangent lifted S 1 -action on TS 2 , (6.1)

J(q 1 , q 2 , q 3 , p1 , p2 , p3 ) = q 1 p2 − q 2 p1 ,

and the classical Hamiltonian

(p1 )2 + (p2 )2 + (p3 )2 + q3, 2 the sum of the kinetic and potential energy. The momentum map J is not proper because the sequence {(0, 0, 1, n, n, 0)}n∈N ⊂ J −1 (0) ⊂ TS 2 does not contain any convergent subsequence. The Hamiltonian H is proper since H −1 ([a, b]) is a closed subset of the compact subset of TS 2 for which 2(a−1) 6 kpk2 6 2(b+1). Therefore, F is also proper. In this case, F (M ) is depicted in Figure 6.1 and the cartographic invariant of (M, F ) is represented in Figure 6.3; we call it ∆(F ). (6.2)

H(q 1 , q 2 , q 3 , p1 , p2 , p3 ) =

Figure 6.1. Image of of F := (J, H) given by (6.1) and (6.2). The edges are the image of the transversally-elliptic singularities (rank 1), the vertex is the image of the elliptic-elliptic singularity (rank 1), and the dark dot in the interior is the image of the focus-focus singularity (rank 0). All other points are regular (rank 2). There is precisely one elliptic-elliptic singularity at ((0, 0, −1), (0, 0, 0)), one focus-focus singularity at ((0, 0, 1), (0, 0, 0)), and uncountably many transversallyelliptic type singularities. The range F (M ) and the set of critical values of F , which equals its bifurcation set, are given in Figure 6.1. The image under F of the focus-singularity is the point (0, 1). The image under F of the ellipticelliptic singularity is the point (0, −1). We know that the image by J of critical points of F of rank zero is the singleton {0}. Hence (one of the two representatives of) ∆(F ) has no vertex in both regions J < 0 and J > 0. In each of these regions, there is only one connected family of transversally elliptic singular values. This means that ∆(F ) in these region consist of a single (semi-infinite)

18

´ ALVARO PELAYO

TUDOR S. RATIU

˜ NGO SAN VU .C

Figure 6.2. Fiber of F := (J, H) given by (6.1) and (6.2) over the focus-focus critical value (0, 1). edge. We can arbitrarily assume that, in the region J < 0, the edge in question is the negative real axis {(y, 0) | y < 0}. Then we have a vertex at the origin (x = 0, y = 0).

Figure 6.3. One of the two cartographic projections of the spherical pendulum. We still need to compute the slope of the edge corresponding to the region where J > 0. For this, we apply [Vu07, Theorem 5.3], which states that the change of slope can be deduced from the isotropy weights of the S 1 momentum map J and the monodromy index of the focus-focus point. (We need to include the focus-focus point because its J-value is the same as the J-value of the elliptic-elliptic point.) So we compute these weights now. The vertex of the polygon corresponds to the stable equilibrium at the South Pole of the sphere. We use the variables (q1 , q2 , p1 , p2 ) as canonical coordinates on the tangent plane to the South Pole. In these coordinates, the quadratic approximation of J is in fact exact, and equal to J (2) = q 1 p2 − q 2 p1 . Now consider the following change of coordinates: q2 − q1 p1 + p2 p1 − p2 q1 + q2 x1 := √ , x2 := √ , ξ1 := √ , ξ2 := √ . (6.3) 2 2 2 2 This is a canonical transformation and the expression of J (2) in these variables is J (2) = 21 (x22 + ξ22 ) − 12 (x21 + ξ12 ). Since the Hamiltonian flows of 21 (x22 + ξ22 ) and 1 2 2 2 (x1 + ξ1 ) are 2π-periodic, this formula implies that the isotropy weights of J at this critical point are −1 and 1. From [Vu07], we know that the difference between the slope of the edge in J > 0 and the slope of the edge in J < 0 must be equal to −1 ab + k, where a and b are the isotropy weights, and k is the


19

monodromy index. For the spherical pendulum, k = 1 because there is only one simple focus-focus point. Thus the new slope is −1 ab + k = 1 + 1 = 2. This leads to the polygonal set depicted in Figure 6.3. 7. Proof of Theorem E We give here the outline of the construction of a family of integrable systems defined on an open subset of S 2 × S 2 , leading to the proof of Theorem E. Step 1. (Construction of suitable smooth functions.) Let Ω := [−1, 1] × [−1, 1] \ {0} × [0, 1].

Let χ : [−1, 1] → R be any C∞ -smooth function such that χ(z2 ) ≡ 1 if z2 ≤ 0 and 0 < χ(z2 ) 6= 1 if z2 > 0. Define f : Ω → R by 1 if z1 ≤ 0; (7.1) f (z1 , z2 ) = χ(z2 ) if z1 > 0.

and note that it is smooth on Ω. Step 2. (Definition of a connected smooth 4-manifold M .) Let S 2 be the unit sphere in R3 and M := S 2 × S 2 \ {((x1 , y1 , z1 ), (x2 , y2 , z2 )) ∈ S 2 × S 2 | z1 = 0, z2 ≥ 0}, where a point in the first sphere has coordinates (x1 , y1 , z1 ) and a point in the second sphere has coordinates (x1 , y2 , z2 ). Since M ⊂ S 2 × S 2 is an open subset, it is a smooth manifold. Moreover, M is connected. Step 3. (Definition of a smooth 2-form ω ∈ Ω2 (M ).) Let πi : S 2 × S 2 → S 2 be the projection on the ith copy of S 2 , i = 1, 2. Let ωi := π ∗ ωS 2 where ωS 2 is the standard area form on S 2 . Define the 2-form ω on M by (7.2)

ω(m1 , m2 ) := (ω1 )m2 + f (z1 , z2 ) (ω2 )m2

for every (m1 , m2 ) ∈ M . Since f is smooth by Step 1, ω is also smooth, i.e., ω ∈ Ω2 (M ). Step 4. (The 2-form ω is symplectic.) One can check that ω is closed because ∂f ∂z1 = 0, and that ω is non-degenerate because f 6= 0.

Step 5. ((M, ω) with J := z1 , H := z2 satisfies {J, H} = 0 and J is a momentum map for a Hamiltonian S 1 -action.) We let S 1 act on M by rotation about the (vertical) z1 -axis of the first sphere and trivially on the second sphere. The infinitesimal generator of this action equals the vector field X ((x1 , y1 , z1 ), (x2 , y2 , z2 )) = ((−y1 , x1 , 0), (0, 0, 0)). This immediately shows that J = z1 is a momentum map for this action. Step 6. ((M, ω) with J := z1 , H := z2 is a generalized semitoric system with only elliptic singularities.) A direct verification shows that the rank zero critical points are precisely (N1 , N2 ), (N1 , S2 ), (S1 , N2 ), and (S1 , S2 ), where Ni , Si are the North and South Poles on the first and second spheres, respectively. One can verify that these critical points are non-degenerate, in the sense that a generic combination (ia, −ia, ibf (1), −ibf (1)) of the linearizations of the vector fields XJ and XH at each of these points (0, 0, if (1), −if (1)) and (0, 0, i, −i) has four distinct eigenvalues. Thus these singularities are of elliptic-elliptic type. The rank one critical points are (N1 , (x2 , y2 , z2 )), (S1 , (x2 , y2 , z2 )), ((x1 , y1 , z1 ), N2 ) with z1 6= 0, and ((x1 , y1 , z1 ), S2 ). Another simple computation shows that all

20

´ ALVARO PELAYO

TUDOR S. RATIU

˜ NGO SAN VU .C

of them are non-degenerate and of transversally elliptic type. It follows that J := z1 , H := z2 is an integrable system with only non-degenerate singularities, of either elliptic-elliptic or transversally elliptic type. Hence (J := z1 , H := z2 ) is a generalized semitoric system. Since the range of F is (7.3)

F (M ) = [−1, 1] × [−1, 1] \ {z1 = 0, z2 ≥ 0},

is not a closed set (see also Figure 7.1), it follows that F is not a proper map.

111111111111 000000000000 000000000000 111111111111 000000000000 111111111111 000000000000 111111111111 000000000000 111111111111 000000000000 111111111111 000000000000 111111111111 000000000000 111111111111 000000000000 111111111111 000000000000 111111111111 000000000000 111111111111 000000000000 111111111111 000000000000 111111111111 Figure 7.1. The image F (M ). Step 7. (Modify F suitably to turn it into a proper map which still defines a semitoric system.) Consider the smooth function g : Ω → R2 defined by 2 +2 , where h(z2 ) > 0, h(z2 ) = 0 if and only if z2 > 0, and g(z1 , z2 ) = z1 , z 2z+h(z 2) 1 h0 (z2 ) < 0 for z2 < 0. Define Fe := F ◦ g = J, J 2H+2 : M → R2 . Since the +h(H) Jacobian of Fe is (z12

1 z12 + h(z2 ) − h0 (z2 )(z2 + 2) > 0 2 + h(z2 ))

(recall that h0 (z2 ) 6 0 and z12 + h(z2 ) > 0 for (z1 , z2 ∈ Ω), it follows that Fe is a local diffeomorphism. In order to show that Fe is proper, it suffices to prove that Fe−1 (K1 × K2 ) is compact if K1 and K2 are closed intervals of R; since the second component of g is always positive, we can assume, without loss of generality, that K2 = [a, b] with a > 0. To show that Fe is proper, we begin by analyzing g −1 (K1 × K2 ). We have (z1 , z2 ) ∈ g −1 (K1 × K2 ) if and only if 2 +2 z1 ∈ K1 and 0 < a 6 z 2z+h(z 6 b, which is implies that ) 1

2

1 z2 + 2 6 6 z12 + h(z2 ). b b Hence either z12 > 1/2b or h(z2 ) > 1/2b. Thus the set g −1 (K1 × K2 ) lies inside the set Ωb in Figure 7.2. Since g −1 (K1 × K2 ) is closed and obviously bounded, as a subset of the compact set Ωb , it follows that g −1 (K1 × K2 ) is compact in R2 . Therefore, Fe−1 (K1 × K2 ) = F −1 g −1 (K1 × K2 )


(−1, 1)

! √ 2 −√ ,1 b

! √ 2 √ ,1 b

(1, 1)

11111111111111111111 00000000000000000000 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 (0, z20 ) 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 (−1, −1)

21

(1, −1)

Figure 7.2. The set Ωb , where z20 < 0 is uniquely determined by the condition h(z20 ) = 1/2b. is compact in S 2 × S 2 and is obviously contained in M , by construction. We conclude that Fe−1 (K1 × K2 ) is compact in M , endowed with the subspace topology. Note that J is not proper because J −1 (0) is not compact. However, Fe is a general semitoric system and Fe is proper. Step 8. (Finding the image Fe(M ).) Let X := [−1, 0) × [−1, 1] ∪ (0, 1] × [−1, 1] ∪ {0} × [−1, 0) . It follows from (7.3) (see also Figure 7.1) that z2 + 2 e (z1 , z2 ) ∈ X . F (M ) = g(F (M )) = z1 , 2 z1 + h(z2 ) Note that the second component of g is an even function of z1 and hence the range Fe(M ) is symmetric about the vertical axis in R2 . A straightforward analysis shows that Fe(M ) is the following region in R2 : 1 3 [ 1 2 6y6 2 {0} × ,∞ ; (x, y) ∈ R 0 < |x| 6 1, 2 x + h(−1) x h(−1)

see Figure 7.3. Note that the closed segment [−1, 1] × {−1} ⊂ F (M ) is mapped by g to the lower curve in Figure 7.3, the two half-open segments ([−1, 1] \ {0}) × {1} to the two upper curves, the two closed vertical segments to the two closed vertical segments, and the half-open interval {0} × [−1, 0) to the infinite halfopen interval {0} × [1/h(−1), ∞). Step 9. (Construction of the cartographic representation.) We shall construct the cartographic invariant in Theorem C from Fe(M ) by flattening out the horizontal curves and setting the height between them at the value given by the volume of the corresponding reduced phase space. For each |x| 6 1, let `(x) denote the volume of the reduced manifold J −1 (x)/S 1 . Then, by Theorem C, the cartographic invariant associated to the general semitoric system (M, Fe) is given by the formula ∆ = (x, y) ∈ R2 | 0 < |x| 6 1, 0 6 y 6 `(x) ∪ {0} × [0, 2π).

22

´ ALVARO PELAYO

˜ NGO SAN VU .C

TUDOR S. RATIU

11111111111111111111 00000000000000000000 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 3 00000000000000000000 y =11111111111111111111 x 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 9 8 7 6

2

5 4 3 2 1 0

−1

−0.5

y=

0

0.5

1

1 x2 +h(−1)

Figure 7.3. The set Fe(M ) with the choice h(−1) = 1.

Using the definition (7.1) of f , a direct computation shows that if x < 0 then J −1 (x)/S 1 = {x} × S 2 , and hence Z Z 1 `(x) = f (x, z2 )dθ ∧ dz2 = 2π f (x, z2 )dz2 = 4π, S2

−1

because for x < 0, we have f (x, z2 ) = 1 for any z2 ∈ [−1, 1]. Similarly, if x > 0 then, as before, the reduced space is J −1 (x)/S 1 = {x} × S 2 , and hence R1 `(x) = 2π 1 χ(z2 )dz2 . If x = 0, then the reduced space J −1 (0)/S 1 is the southern hemisphere of the second factor and hence `(0) = 2π. Therefore, the cartographic invariant is given in Figure 7.4. Z 11111111111111111111 00000000000000000000 χ(z )dz 2π 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 4π 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 2π 11111111111111111111 00000000000000000000 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 (−1, 0) (1, 0) 1

2

2

−1

x

Figure 7.4. A representative of ∆(M, Fe).

We have so far shown (E.1)-(E.5). Theorem D implies (E.6). We have left to show (E.7). To conclude the proof, we modify the construction above in order to illustrate the existence of unbounded cartographic invariants with fibers of infinite length. As we shall see, most of the computations of the previous example remain valid.


23

Let

N := S 2 × S 2 \ {((x1 , y1 , z1 ), (x2 , y2 , z2 )) ∈ S 2 × S 2 | z1 = 0, z2 > 0}

(7.4)

∪ {((x1 , y1 , z1 ), (x2 , y2 , z2 )) ∈ S 2 × S 2 | z1 > 0, z2 = 1} .

As in the previous example, N is open and connected. Moreover, because it is a subset of M , the restriction of the form Ω given by (7.2), is a symplectic form. Similarly, J = z1 , H = z2 defines an integrable system on N and J is the momentum map of a Hamiltonian S 1 -action. The computations in the previous example show that we have the same singularities, all of them non-degenerate. If F = (J, H), its image is (7.5) F (N ) = [−1, 1] × [−1, 1] \ {z1 = 0, z2 ≥ 0} ∪ {0 6 z1 6 1, z2 = 1} (see Figure 7.5) which is not a closed set, and hence F is not a proper map. Define

111111111111 000000000000 000000000000 111111111111 000000000000 111111111111 000000000000 111111111111 000000000000 111111111111 000000000000 111111111111 000000000000 111111111111 000000000000 111111111111 000000000000 111111111111 000000000000 111111111111 000000000000 111111111111 000000000000 111111111111 000000000000 111111111111

Figure 7.5. The image F (N ).

g(z1 , z2 ) :=

z2 + 2 z1 , 2 ((z1 − 1) + h(z1 ))(z12 + h(z2 ))

and Fe := g ◦F , where F := (z1 , z2 ); h is as in the previous example. To see that g is a local diffeomorphism, it sufficesto note that the Jacobian determinant of ∂∆ g has the expression ∆ − (z2 + 2) ∂z /∆2 , where ∆ := ((z1 − 1)2 + h(z1 ))(z12 + 2 h(z2 )). Since ∆ > 0 and ∂∆/∂z2 = 2(z2 − 1)(z12 + h(z2 )) + ((z1 − 1)2 + h(z1 ))h0 (z2 ) < 0, it follows that the Jacobian determinant of g is strictly positive. As in the previous example, one can check that g −1 (K1 × K2 ) is a compact subset of R2 , where Ki , i = 1, 2, are closed bounded intervals in R. The argument given in the previous example shows then that Fe is a proper map. Therefore, (N, Fe) is a proper general semitoric system. The image Fe is given in Figure 7.6. Finally, to determine the possible affine invariants associated to this system, we need to compute `(x), the volume of the reduced manifold J −1 (x)/S 1 . As

´ ALVARO PELAYO

24

TUDOR S. RATIU

˜ NGO SAN VU .C

11111111111111111111111 00000000000000000000000 y 00000000000000000000000 11111111111111111111111 00000000000000000000000 11111111111111111111111 00000000000000000000000 11111111111111111111111 00000000000000000000000 11111111111111111111111 00000000000000000000000 11111111111111111111111 00000000000000000000000 11111111111111111111111 3 y = x h(x) 00000000000000000000000 11111111111111111111111 00000000000000000000000 11111111111111111111111 00000000000000000000000 11111111111111111111111 F˜ (M ) 00000000000000000000000 11111111111111111111111 00000000000000000000000 11111111111111111111111 00000000000000000000000 11111111111111111111111 00000000000000000000000 11111111111111111111111 00000000000000000000000 11111111111111111111111 00000000000000000000000 11111111111111111111111 00000000000000000000000 11111111111111111111111 00000000000000000000000 11111111111111111111111 00000000000000000000000 11111111111111111111111 00000000000000000000000 11111111111111111111111 00000000000000000000000 11111111111111111111111 00000000000000000000000 11111111111111111111111 00000000000000000000000 11111111111111111111111 2

−1

0 y=

1

x

1 (4+h(x))(x2 +1)

Figure 7.6. The image Fe(N ) with the choice h(−1) = 1.

before, we compute

`(x) =

  

4π, 2π,

  2π(1 + α),

if x < 0 if x = 0 if x > 0

R1 where α := 0 χ(z2 )dz2 > 0. The possible cartographic invariants are given in Figure 7.7. This proves (E.7). α=0 4π 2π

4π 2π

11111111111111111111 00000000000000000000 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 11111111111111111111 00000000000000000000 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 00000000000000000000 11111111111111111111 1≤α